ARTICLE IN PRESS

Journal of the Mechanics and Physics of Solids 55 (2007) 26722686 www.elsevier.com/locate/jmps

Shock enhancement of cellular structures under impact loading: Part II analysis

S. Pattofattoa, I. Elnasria, H. Zhaoa,, H. Tsitsirisa, F. Hilda, Y. Girardb
a

canique et Technologie (LMT-Cachan), ENS-Cachan/CNRS-UMR8535/Universite Paris 6, Laboratoire de Me sident Wilson, F-94235 Cachan Cedex, France 61 avenue du Pre b EADS-CCR Suresnes, 12 bis rue Pasteur, F-92152 Suresnes Cedex, France Received 25 January 2007; received in revised form 4 April 2007; accepted 8 April 2007

Abstract Numerical simulations of two distinct testing congurations using a Hopkinson bar (pressure bar behind/ahead of the shock front) are performed with an explicit nite element code. It allows us to conrm the observed test data such as velocity and force time histories at the measurement surface. A comparison of the simulated local strain elds during shock front propagation with those measured by image correlation provides an additional proof of the validity of such simulations. Very simple rate insensitive phenomenological constitutive model are used in such simulations. It shows that the shock effect is captured numerically with a basic densication feature. It means that strength enhancement due to shock should not be integrated in the constitutive model of foam-like materials used in industrial FE codes. In order to separate shock enhancement from entire strength enhancement, an improvement of an existing model with easily identiable parameters for shock enhancement prediction is proposed. For a quick estimate of the shock enhancement level, a simple power law densication model is proposed instead of the classical RPPL model proposed by Reid and co-workers [Tan et al., 2005. Dynamic compressive strength properties of aluminium foams. Part Iexperimental data and observations. J. Mech. Phys. Solids 53, 21742205]. It is aimed at eliminating the parameter identication uncertainty of the RPPL model. Such an improved model is easily identiable and gives a good prediction of the shock enhancement level. r 2007 Elsevier Ltd. All rights reserved.
Keywords: Shock wave; Foam; Cellular materials; Impact; Numerical simulation

Corresponding author.

E-mail address: zhao@lmt.ens-cachan.fr (H. Zhao). 0022-5096/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.jmps.2007.04.004

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2673

1. Introduction The concept of shock enhancement effect under high speed impact (4100 m/s) was originally proposed by Reid and Peng (1997) to explain testing results on woods. Afterwards, a number of authors also reported this effect for various cellular materials at high impact speeds (Lopatnikov et al., 2003, 2004; Tan et al., 2002, 2005; Radford et al., 2005). For relatively low impact speeds around the so-called critical velocity under which shock enhancement is not signicant ($50 m/s), shock enhancement is experimentally studied with a 60-mm diameter Nylon Hopkinson bar (see companion paper, Elnasri et al., 2007). With a single bar, tests with two congurations using a large diameter soft Hopkinson bar behind/ahead of the shock front allow for the estimation of the stress jump across the shock front as well as the shock front speed. Such tests show a signicant shock enhancement for two materials (namely, Alporas foam and hollow spheres). Experimental data in previous works prove then the existence of such shock effect (Tan et al., 2002, 2005; Elnasri et al., 2007). However, only global measurements are available (velocity/force in the pressure bar, or surface strain maps). For a better understanding of the shock enhancement mechanism, a numerical study is a complementary means. It provides all the virtual details that are difcult to measure in a real test. It will also help us conrm which model characteristics of foam-like materials are responsible for this shock enhancement. Besides, it is also important to determine how to deal with such enhancement in industrial applications. For example, is it necessary or not to introduce the shock enhancement effect in the material constitutive law, especially for impact speeds around the so-called critical velocity where the shock effect is of the same magnitude as materials rate sensitivity. For this purpose, numerical analyses for the two testing congurations using LS-Dyna explicit nite element code with a macroscopic constitutive law (crushable foam) are performed. They show that such a simple rate-insensitive constitutive model is able to reproduce the essential features of shock enhancement. It may be concluded that (i) the numerical shock enhancement depends only on a simple macroscopic densication constitutive law, and (ii) the nature of the microstructure of cellular materials has no signicant inuence. Therefore, shock enhancement should be eliminated from the entire observed enhancement for the constitutive law development. A simple prediction is also needed. For example, the rigid perfectly plastic locking (RPPL) model proposed by Reid and Peng (1997) gives a fast estimation. However, the shock stress jump and shock front speed predicted by the RPRL model are too sensitive to the parameters (e.g., the rigid locking strain). In addition, these parameters are difcult to choose. It leads us to propose another model that assumes a power law densication. Such a model enables for the elimination of the uncertainty induced by the arbitrary choice of the rigid locking strain in the RPPL model and it gives a closed-form expression and a good prediction. 2. Numerical analysis Numerical analyses are performed to have access to local values that cannot be measured in real tests. It is also aimed at verifying if the shock enhancement depends only on the macroscopic phenomenological feature or may depend also on the microstructure

ARTICLE IN PRESS
2674 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

of cellular materials. If a simple phenomenological foam-like material behaviour is sufcient to reproduce all the shock enhancement features, it means that the microstructure will have a very limited inuence on shock enhancement. 2.1. FE model The LS-Dyna explicit nite element code is used to perform the simulations. The specimen whose diameter is 60 mm and thickness 40 mm (Fig. 1) is meshed with 32,000 cubic 8-node elements (with 20 elements in the length). The constitutive law of the foam is given by the so-called crushable foam model available in the LS-Dyna code (Hallquist, 1998). At time step n, the crushable foam model denes at rst a trial stress tensor sn1 trial at time step n+1, based on an elastic prediction ij sn1 trial sn E_n Dtn , ij ij ij sn , n , ij _ ij (1) Dtn are, respectively, the stress tensor, the strain rate and time step; E is the where elastic constant. Afterwards, the principal stresses of this trial stress tensor strial , k 1; 3 are compared to k the yield stress sy. If jstrial j4sy , the principal stress is scaled back by sin1 sy strial =jstrial j. i i i The new stress tensor sn1 is nally obtained by transforming scaled principal stresses to ij the initial coordinate system. The yield stress sy is considered as a function of volumetric strain ev v 1 V , V0 (2)

where V is the volume and V0 the initial volume of an element. From experimental nominal stressstrain curves, it is straight forward to derive the volumetric strain ev from nominal strain enominal, given that the Poissons ratio is set to zero (Zhao et al., 2005) v 1 L nominal . L0 (3)

Fig. 1. Finite element mesh of the specimen.

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2675

Fig. 2 shows the ow stress-volumetric strain relationship used in the following simulations (dened by discrete points), which is derived from a compression test on Alporas foam (Elnasri et al., 2007). 2.2. Numerical simulation for the two test congurations The two test congurations presented in a companion paper (Elnasri et al., 2007) are now analysed. 2.2.1. Conguration 1 For test conguration 1 where the projectile strikes the foam specimen cemented on a 60-mm nylon Hopkinson pressure bar, the measured velocities are prescribed at both sides of the specimen by means of two moving rigid walls (Fig. 3). The velocities of the rigid walls are shown in Fig. 4. The comparison between experimental and numerical results is carried out by computing the force ahead of the shock front measured by the Hopkinson bar. The good agreement

Fig. 2. Input ow stress-volumetric strain relation for Alporas foam.

Fig. 3. Boundary conditions for simulating test conguration 1.

ARTICLE IN PRESS
2676 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

Fig. 4. Input and output prescribed velocities for the moving rigid walls (cong. 1).

Fig. 5. Comparison between simulated and measured force/time histories (cong. 1).

between simulated and measured force/time histories (Fig. 5) validates the FE model and the constitutive law. 2.2.2. Conguration 2 For test conguration 2 where the foam specimen is cemented on the projectile and strikes the Hopkinson pressure bar, the pressure bar is represented by a moving wall travelling at the measured velocity, in the same way as for conguration 1. However, only the measured initial velocity is prescribed to the foam specimen and to the projectile considered as a rigid body (Fig. 6). Here the mass of the real projectile used in the test is associated with the numerical rigid body projectile model. The comparison is also performed by considering the force measured by the Hopkinson bar (i.e., behind the shock front). Fig. 7 shows a good agreement between simulated and measured forces.

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2677

Fig. 6. Foam and projectile against a moving rigid wall (cong. 2).

Fig. 7. Comparison between simulated and measured force/time histories (cong. 2).

2.3. Discussions on shock enhancement simulation The simulations of the two loading congurations provide a satisfactory agreement between the simulated and measured forces by Hopkinson bars. Since these two testing congurations are used to measure the force jump across the shock front, it is interesting to study the two forces across the shock front for a simulated test. If one takes the simulation of a test in conguration 1 as an illustration example, the simulated forces ahead of and behind the shock front are plotted in Fig. 8. In Fig. 8a, one notes that the history is similar to the measured forces between two testing congurations at 55 m/s (Elnasri et al., 2007). In Fig. 8b, a simulated test at 19 m/s with the same constitutive model is shown, where no shock effect is observed. A quantitative comparison between the simulated forces across the shock front in conguration 1 (see Fig. 8a) and the two measured forces, supposed to be those across the shock front, obtained by two tests under different congurations is given in Fig. 9. One concludes that the measured and simulated forces are close. This proves that the concept of testing with two congurations does provide a means of measuring simultaneously the forces at both sides and can be used to investigate shock enhancement.

ARTICLE IN PRESS
2678 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

Fig. 8. Simulated forces ahead of and behind the shock front (cong. 1): (a) impact velocity: 55 m/s; (b) impact velocity: 19 m/s.

Let us now study the details of the strain discontinuity propagation. The simulated strain map at different instants is also compared with the strain eld obtained by image correlation (see companion paper). The simulated average strain evaluated in the same manner as by image correlation post-processing (dotted line, Fig. 10) is compared to the experimental data (solid line, Fig. 10). A good agreement is obtained. With this simulation method, a prediction of stress jump for different materials at any given impact velocity is obtained. Table 1 gives a prediction of the stress enhancement ratio across shock font for Alporas foam up to 200 m/s. Such prediction at 55 m/s corresponds to that experimentally measured at 55 m/s. This ratio for Cymat foam shows a huge shock enhancement ratio for high impact speed (see also Tan et al., 2005), even though no shock front is clearly measured at 45 m/s by our device.

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2679

Fig. 9. Comparison between experimental and simulated force histories.

Fig. 10. Comparison of simulated (FE) and measured (CP) strain discontinuity proles.

Table 1 Simulated shock enhancement for Alporas and Cymat foams Speed (m/s) sshock sy =sy Alporas 56 100 150 200 0.44 1.4 3.2 5.4 Speed (m/s) 44 100 150 200 sshock sy =sy Cymat 0.20 0.63 1.45 2.68

ARTICLE IN PRESS
2680 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

3. Numerical shock enhancement and RPRL shock model In the previous simulations, the foam is represented by a rate insensitive and macroscopic constitutive law. Simulations for other tested materials such as hollow spheres, Cymat foams and honeycombs are also performed with the same numerical procedure. The only difference is the global phenomenological law used (Fig. 11). It means that the microscopic nature has no signicance for shock enhancement, provided the macroscopic nominal stressstrain curve is identical. It also means that shock enhancement is a structural effect that is numerically reproducible for any concave stressstrain relationship. Therefore, it should not be considered as an intrinsic part of the material behaviour and should not be integrated into the constitutive law. To eliminate such shock enhancement, a simple rule is to ensure that the measuring system is always placed ahead of the shock front during the test. It is unfortunately not always possible because of the used experimental set-up. Consequently, a simple shock model should be used to provide an acceptable prediction. Let us consider the RPPL model. Following the same denitions as those used before (see companion paper), r0 , sy, lock denote the initial density, the plastic ow stress and the locking strain, and Vimpact, sshock, U the initial impact velocity, the stress behind the shock front and the shock front velocity, respectively. With the simplications of the RPPL model, the shock front velocity and stress behind the shock front are calculated as U V impact ; lock sshock sy r0 V 2 impact . lock (4)

Arbitrary identication is carried out for the average ow stress and the locking strain from the nominal stressstrain curve (Fig. 11). Identied values of the basic mechanical parameters for the studied materials are given in Table 2. The ow stress in Table 2 is dened as the average plateau stress under quasi-static loading. The locking strain is dened visually.

Fig. 11. Typical quasi-static behaviour of tested materials.

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 Table 2 Basic mechanical data of the studied materials Materials Alporas Hollow spheres Cymat foam Honeycomb Density (kg/m3) 245 219 235 38 Flow stress (MPa) 1.8 1.1 3.8 0.6 Locking strain 0.7 0.7 0.6 0.8 2681

Table 3 Comparison between experimental data and RPRL predictions Materials r0 lock sy 1.80 104 2.84 104 1.03 104 0.79 104 sshock sy =sy experiment 0.44 0.66 0. 0. sshock sy =sy model 0.54 0.86 0.21 0.15 U (m/s) experiment 94 (87) 107 U (m/s) model 78 78

Alporas (55 m/s) Hollow sphere (55 m/s) Cymat foam (45 m/s) Honeycomb (45 m/s)

Eq. (4) shows that the shock enhancement ratio sshock sy =sy is determined by the ratio between the density and the ow stress multiplied by the locking strain r0 =lock sy . An increase of this value yields an increase of the shock enhancement ratio. A comparison between the RPPL model prediction and the experimentally measured shock enhancement is given in Table 3. Such model may explain why Alporas foams as well as hollow spheres have a more signicant enhancement, whereas Cymat foams and honeycombs exhibit almost no effect. One important remark is that Cymat foams as well as honeycombs would also exhibit a shock effect for high impact velocities, which was observed for Cymat foams (Tan et al., 2002). The present simulations on honeycombs show also shock enhancement at higher impact speeds. 4. Power law shock model 4.1. Parameter sensitivity of RPPL model From the aforementioned predictions (Table 3), it is shown that the RPPL model does provide a rst-order approximation. However, the identication of the parameters of the model is not easy. The plastic ow stress sy is more or less easy to identify, the value of the locking strain is arbitrary due to the non-linear shape of the densication curve. The shock front velocity and the stress behind the shock front are very sensitive to this locking strain. Fig. 12 shows the change of these values predicted by the RPPL model with the locking strain for ve different initial impact velocities (20, 40, 80, 200 and 300 m/s). The difference is really signicant for different values of the locking strain, especially for high impact velocities. To overcome this identication difculty, we propose to account for the progressive densication feature. The same impact case as discussed in Eq. (4) is considered. The fact that the densication is progressive implies that the strain reached behind the shock front

ARTICLE IN PRESS
2682 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

Fig. 12. Dependence of the shock front velocity (a) and shock stress (b) with the locking strain for the RPRL model when ve impact velocities (in m/s) are considered.

eshock is not explicitly known as the locking strain in the RPPL model. The basic continuity equations (4) are replaced by U V impact shock and sshock sy r0 V 2 impact shock (5a)

and the solution is dened by using the stressstrain relation of the material sshock f shock . (5b)

Considering the case of Alporas foams, where the stressstrain curve is given in Fig. 13, a numerical solution is obtained from Eqs. (5a) and (5b). Using a polynomial approximation of relation (5b) obtained from experimental stressstrain relationships (Fig. 11) with a curve-tting technique, on evaluates numerically the strain behind the shock front eshock with respect to the impact velocity. The numerical result of the change of this value with respect to the impact velocity is plotted in Fig. 13 where the solid line is the curve t of experimental

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2683

Fig. 13. Change of shock strain for ve values of the impact velocity (in m/s).

data (points). The circles in this gure show the shock strain and stress behind the shock front for different impact velocities. One notes that the value of the shock strain eshock varies in a wide range, namely, from 60% for an impact velocity of 50 m/s, to 90% at 300 m/s. 4.2. An improved shock model with a hardening locking The aforementioned analysis shows the need for improving the RPPL model. A numerical solution may be obtained by solving Eqs. (5). However, for a simple use, an explicit closed-form solution is preferred. We propose to introduce a model using a power law without locking strain to replace the RPPL model. The densication curve is dened by the initial yield stress sy, the power m, and the coefcient k s sy km . Eqs. (5a) and (6) then lead to km shock r0 V 2 impact . shock (7) (6)

The derivation of the shock strain eshock is straight forward !1=m1 r0 V 2 impact shock k and the shock front velocity reads !1=m1 kV m1 impact U . r0

(8)

(9)

From experimental stressstrain data, it is easy to identify such model. Table 4 provides the parameters for the four studied materials. The quality of tuning for Alporas foam is shown in Fig. 14 as an example.

ARTICLE IN PRESS
2684 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

Fig. 15 gives the change of the shock front velocity with the impact velocity for different models of the stressstrain curve. The dashed lines are the results that would be given by the RPPL for two limiting values of the locking strain, namely, 60% and 80%. It shows that the explicit solution of a power law model gives a satisfactory prediction.
Table 4 Identied parameters for power law locking Material Alporas Hollow sphere Cymat foam Honeycomb sy (MPa) 1.70 0.97 3.97 0.671 m 6.40 3.536 8.99 38.19 K (MPa) 21.2 8.48 48.3 653

Fig. 14. Power law model identication for Alporas foam.

Fig. 15. Comparison of models in terms of shock front velocity vs. impact velocity.

ARTICLE IN PRESS
S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686 2685

Fig. 16. Shock enhancement effect for the four studied materials.

Fig. 17. Comparison between FEM analyses and power model for the prediction of the relative shock stress vs. impact velocity.

From this power law model with identied parameters (Table 4), the relative shock stress jump sshock sy =sy is also obtained. The shock enhancement effect for those four materials is plotted in Fig. 16. The comparison between FEM simulation results for Alporas and Cymat foams (Table 2) and present model is given in Fig. 17. A reasonable agreement is found. 5. Conclusion In this paper, numerical simulations of tests in two different congurations are performed with an industrial explicit FE code. The results show that shock enhancement of foams is reproduced by numerical simulations using a simple macroscopic, homogenous

ARTICLE IN PRESS
2686 S. Pattofatto et al. / J. Mech. Phys. Solids 55 (2007) 26722686

and rate-insensitive constitutive law based on a nominal stressstrain relationship obtained in a quasi-static compression test. It means that such shock enhancement effect should not be taken into account at the level of the constitutive law itself. Shock enhancement of cellular materials is governed by macroscopic structural parameters. Test results on cellular materials with different base materials and microstructures show that such enhancement is independent of the microstructure and the local crushing deforming mode. Last, the widely used RPPL model is shown to be very sensitive to the identication uncertainty of the constitutive parameters. Since the rigid locking strain is only determined in an arbitrary manner, the RPPL model is not accurate enough, especially for low impact velocities (i.e., less than 100 m/s). Since experimental and numerical results show that the strain behind the shock front depends on the impact velocity, the RPPL model considering this value as a constant (locking strain) cannot be accurate. An improved model based on a power law densication assumption allows for an easy determination of its parameters from experimental data, and gives results in good agreement with experimental data. References
Elnasri, I., Pattofatto S., Zhao, H., Tsitsiris, H., Hild F., Girard, Y., 2007. Shock enhancement of cellular structures under impact loading: part I, Experiments. J. Mech. Phys. Solids, in press, doi:10.1016/ j.jmps.2007.04.005. Hallquist, J.O., 1998. Ls-Dyna Theoretical Manual. Livermore Software Technology Corporation. Lopatnikov, S.L., Gama, B.A., Haque, Md.J., Krauthauser, C., Gillespie, J.W., Guden, M., Hall, I.W., 2003. Dynamics of metal foam deformation during Taylor cylinder-Hopkinson impact experiment. Compos. Struct. 61, 6171. Lopatnikov, S.L., Gama, B.A., Haque, Md.J., Krauthauser, C., Gillespie, J.W., Guden, M., 2004. High-velocity plate impact of metal foams. Int. J. Impact Eng. 30, 421445. Radford, D.D., Deshpande, V.S., Fleck, N.A., 2005. The use of metal foam projectile to simulate shock loading on a structure. Int. J. Impact Eng. 31, 11521171. Reid, S.R., Peng, C., 1997. Dynamic uniaxial crushing of wood. Int. J. Impact Eng. 19, 531570. Tan, P.J., Harrigan, J.J., Reid, S.R., 2002. Inertia effects in uniaxial dynamic compression of a closed cell aluminium alloy foam. Mater. Sci. Technol. 18, 480488. Tan, P.J., Reid, S.R., Harrigan, J.J., Zou, Z., Li, S., 2005. Dynamic compressive strength properties of aluminium foams. Part Iexperimental data and observations. J. Mech. Phys. Solids 53, 21742205. Zhao, H., Nasri, I., Abdennadher, S., 2005. An experimental study on the behaviour under impact loading of metallic cellular materials. Int. J. Mech. Sci. 47, 757774.